【作者】 张磊；

【导师】 吕锡香；

【作者基本信息】 西安电子科技大学 ， 信息安全， 2018， 硕士

【摘要】 随着互联网技术的迅速发展,越来越多的用户热衷于将个人数据存储在网络中以便快捷地访问和处理。然而,近些年云服务器频繁被攻击而引发了各种隐私数据泄露事件,使得人们意识到在追求方便的同时也不能忽视个人数据的隐私安全。为保护隐私,用户会将自己的数据加密后上传到云服务器中进行存储,而不是将隐私和安全保护寄希望于云服务提供商承诺的安全控制系统。但是,这样会导致传统的针对明文的数据处理方法不再适用,对密文难以进行直接的数据处理。因此,如何在密文域上对数据进行处理成为了研究的一大热点,其中包括密文域数据的安全检索,合法性审查以及安全检查等操作。目前,针对如何在密文域上进行数据处理,许多学者和研究人员提出基于同态加密的安全信号处理方案。利用这种技术,可以在用户数据不被解密的同时对用户密文数据进行处理,实现高效的图像安全检索和合法性审查。本论文主要研究密文域下的图像数据处理问题,分别提出了一种基于小波不变矩图像特征的安全检索方案和合法性审查方法。论文的主要贡献如下:(1)小波不变矩是图像处理领域一种重要的图像特征,它结合了小波变换和几何不变矩特征,可以很好的描述图像的全局和局部特征。然而由于小波不变矩的计算涉及到密文域难以处理的复数和小数,使得其在密文域上的计算难以完成。结合Paillier同态加密技术,我们引入了一种可以在密文域上处理复数和小数的算法,提出了密文域图像小波不变矩特征的计算方法。基于该方法提取的密文图像特征在解密后和明文特征相等,能维持较好的图像特征描述。进一步,我们基于提出的密文图像特征计算方法,设计了一种图像的安全检索方案,可以在保护用户检索隐私的条件下,返回给用户存储在服务器端的个人图像。实验证明该方案有很好的性能表现和鲁棒性,云服务器在完成规定的协议时,不能获取任何用户的隐私信息。(2)基于云服务或社交平台的分享功能,非法图片得以在网络上的快速分享和扩散。针对这一问题,利用密文域图像小波不变矩计算方法,我们提出了一种对用户图像进行合法性审查的方案。用户对图像进行加密上传时,由服务器对用户图像在密文域提取图像特征,并计算和数据库中非法图片的特征距离,判别是否是非法图像。该方案可以有效的检测出非法图像,并阻止非法内容在互联网上的传播和扩散。仿真实验结果表明,我们提出的方案有着很好识别率,并且可以保护用户和审查方两端的隐私,具有很好的应用前景。更多还原

【Abstract】 With the rapid development of Internet,users are increasingly interested in storing personal data in cloud for conveniently accessing and processing.However,because the Internet events of privacy leak happened frequently in recent years,some people have noticed that they must take some measures to preserve their privacy when they outsourced their data to cloud.In order to protect the data from attackers,a user encrypts it before uploading it to the cloud server.However,traditional data processing methods based on plaintext,in this case,are not suitable for directly processing the data in ciphertext.So,how to process data in the encrypted domain has been becoming a hot research topic,which includes some applications like secure image retrieve,legitimacy check,safety review and so on.Fortunately,based on homomorphic cryptosystems,scholars and researchers have proposed some algorithms for processing signals in the encrypted domain,some of which can be used to design image retrieve schemes and legitimacy check schemes.Due to the signal processing algorithms in the ciphertext domain,the cloud server can process encrypted data without decrypting it.In this paper,we mainly study the methods for processing image data in the encrypted domain.We propose an algorithm for computing image Wavelet invariant moment in the encrypted domain,and based this algorithm we further give a secure image retrieve scheme and a legality review method.Specifically,the main contributions of the paper are as follows:(1)Wavelet invariant moment,a kind of important image features in the image processing field,combines the wavelet transform and geometric moment,and it can well describe images in both global and local scopes.However,it is difficult to compute the Wavelet invariant moment in the encrypted domain,because the computation involves complex numbers and floating numbers which cannot be processed in homomorphic cryptosystem.In this paper,we introduce an algorithm that enables the Paillier cryptosystem to process the complex numbers,and propose a method to compute image Wavelet invariant moment in the encrypted domain.After decryption,Wavelet invariant moment achieved in ciphertext is equivalent to that computed in plaintext,which maintains good image feature description.Further,with the proposed feature computation scheme in the encrypted domain,we design a secure image retrieve scheme,which enables a user to securely search and retrieve images from his/her personal image library on the cloud.The experiment shows that a server cannot get any user’s information when completing the protocol and also demonstrates that our scheme has a good performance and robustness.(2)By using the photo-sharing function of cloud,a greate many illegal images are shared and diffused on Internet.Against this problem,we propose a scheme to check the legality of users’ images based on the computation method of wavelet invariant moment in the encrypted domain.In this scheme,the user encrypts his image and uploads the encrypted image to the server,and then the server computes the image features as well as the distance between this image and the illegal images in its own database.By the distance,the cloud sever is able to determine whether this image is an illegal one.This scheme can effectively detect illegal images and prevent illegal content from spreading on the Internet.The simulation shows that the proposed scheme has a good recognition rate and can protect the privacy of both the user and the censor,which has a good application prospect.更多还原